While most AI companies are still engaged in intense competition within superficial scenarios like "edge assistants" and "knowledge base Q&A," Hai Zhi Technology has leveraged its unique graph-model fusion technology to truly embed intelligent agents (AI Agents) into the core business processes of countless industries, enabling them to independently undertake high-precision, serious tasks such as analysis, decision-making, and execution. As a rare contender aiming to become the "first AI de-hallucination stock," Hai Zhi Technology has precisely positioned itself in the deep-water zone of industrial-grade intelligent agents, establishing dual barriers in both technology and commercialization, while also capitalizing on the industry's explosive growth trend, positioning itself as a key navigator guiding AI from technological euphoria to real-world industrial implementation.
The core demands of industrial-grade scenarios for AI have always revolved around trustworthiness, controllability, and traceability. However, the inherent hallucination problems of general large language models present a fatal obstacle to their adoption in serious fields such as financial risk control, government decision-making, and power grid and traffic scheduling. According to data from Frost & Sullivan, Hai Zhi Technology is the first company in China to effectively reduce large model hallucinations through knowledge graphs, building core competitiveness and technological barriers across three key dimensions: factual accuracy, hallucination rate, and complex logical reasoning capabilities, successfully mitigating the hallucination issues of large language models.
Unlike the external patch-style error correction used by most vendors, Hai Zhi Technology's graph-model fusion technology addresses the challenge of AI hallucinations at its source. The concept of the knowledge graph, proposed by Google in 2012, is hailed as the "digital brain" of AI. Its typical capabilities include intuitive presentation, hidden relationship mining, deep-level reasoning, and event diffusion analysis. Combining knowledge graphs with large language models can effectively eliminate large model hallucinations. Applying this technology to professional fields relies on the structured logic of "entity-relationships" to systematize and structure scattered, fragmented industry knowledge, ensuring that every AI output has verifiable factual evidence.
Hai Zhi Technology's core competitiveness does not lie solely in innovation within the single field of "knowledge graphs," but rather in a "trinity" of capabilities. First,卓越的图计算能力. Hai Zhi Technology's graph computing capability is based on three key technological innovations: a native graph-based approach for processing large-scale data storage; advanced graph computing techniques utilizing graph reduction and repartitioning; and a unique subgraph cell batch processing technology. These innovations enhance the data throughput of graph databases, improve processing performance, and reduce latency. In 2023, the company's self-developed Atlas Graph database broke a world record with a performance 45% higher than the previous record, achieving the top comprehensive performance score in the Linked Data Benchmark Council test. Frost & Sullivan data indicates that as a high-performance distributed graph database, the company's Atlas Graph supports real-time analysis of trillions of data points, far exceeding the industry average.
Second,图模融合技术. The company integrates large language models and knowledge graphs into multiple stages (including pre-training, inference, and retrieval stages) to achieve deep semantic understanding, traceable and explainable answers, and automated end-to-end reasoning. For instance, by enabling large language models to learn graph reasoning capabilities during pre-training and by absorbing structural knowledge from graph databases, the understanding of implicit relationships by large language models can be improved, effectively reducing hallucinations.
Third,应用专长. Over the past decade, the company has accumulated application expertise in enterprise data governance, analysis, and algorithm solutions covering multiple scenarios. This rich experience has provided a deep understanding of scenario-specific data and enabled the construction of a universal knowledge framework tailored to different application scenarios.
To use a通俗的 analogy for Hai Zhi Technology's core technology, it is like solving the decades-long "left-brain/right-brain coordination" problem in industrial AI. The symbolism represented by knowledge graphs is like AI's "left brain," excelling at rational reasoning, logical verification, and factual storage. The connectionism represented by large language models is akin to AI's "right brain," skilled at semantic understanding, divergent association, and efficient interaction. These two techniques have alternately had their moments of glory in AI history: symbolism once defeated chess players with clear logical rules, while connectionism broke through by defeating Go champions with deep learning, yet each has limitations when operating alone.
Hai Zhi Technology integrates both symbolism and connectionism throughout the entire workflow of the Atlas intelligent agent. For example, during the pre-training phase, the large model learns the structured logic of the knowledge graph, establishing factual cognition in advance. During the inference phase, the large model is responsible for divergent association and understanding complex requirements, while the knowledge graph simultaneously performs logical verification, ensuring every conclusion is supported by factual "entity-relationship" evidence. This deeply integrated, rather than simply superimposed, technical approach is the key barrier that distinguishes Hai Zhi from other industrial AI vendors and has allowed it to capture a dominant share of the graph-centric AI intelligent agent赛道.
The ultimate value of graph-model fusion technology must be realized through implementation in industrial scenarios. Hai Zhi Technology's commercial confidence stems from its solutions' ability to genuinely address core customer pain points, making customers willing to pay for high-value services. According to an earlier Financial Times report, in the financial sector, the company's personal credit application anti-fraud relationship graph project for Bank of Shanghai led to a 20% month-on-month increase in independently discovered syndicated fraud risk events in credit card scenarios and reduced risk losses by hundreds of millions of yuan annually in the personal loan scenario. The enterprise-level knowledge graph system built for Guangdong Huaxing Bank successfully implemented anti-money laundering suspicious transaction monitoring, winning the first prize for typical practice from the Guangzhou branch of the People's Bank of China due to its outstanding accuracy. These tasks are directly related to banks' risk control and compliance operations, representing the very lifeblood of their business.
Public information gathered shows that numerous large and medium-sized domestic financial institutions, including Bank of Communications, Bank of China, China Merchants Bank, CITIC Bank, and Qingdao Bank, are among the company's clients. Based on available public information, in the government and security sectors, Hai Zhi's intelligent agents have achieved cross-departmental data integration and traceable intelligent analysis. In scenarios such as anti-fraud and regulatory inspections, they accurately identify illegal activities through complex relationship network mining, effectively replacing traditional manual inspection methods and achieving dual improvements in both efficiency and accuracy.
Breakthroughs have also been made in the energy and industrial manufacturing sectors. Participating in power grid intelligent upgrades, the company uses graph-model fusion technology to analyze equipment correlation relationships and operational data, achieving core operations like fault prediction and dispatch optimization, thereby ensuring stable equipment operation while significantly reducing operational costs. These implementation cases collectively demonstrate that Hai Zhi's intelligent agents are not merely tools to assist human labor; they are core engines capable of independently undertaking serious business tasks and driving industrial digital transformation.
According to the prospectus, the company currently serves over 360 government and enterprise clients, covering sectors with strong payment capabilities such as government affairs, finance, and telecommunications. The average contract value is at a relatively high industry level, forming a virtuous cycle of "having technological barriers and willing customers."
The industrial-grade intelligent agent market naturally has分层 of "shallow waters" and "deep waters." The shallow waters focus on low-precision scenarios like general Q&A and process assistance, characterized by low entry barriers, intense competition, and limited value. The deep waters target scenarios involving core business decision-making and high-precision execution, imposing extremely high requirements for technological barriers, industry knowledge reserves, and compliance capabilities, making them inaccessible to ordinary vendors. Hai Zhi Technology's赛道选择 has avoided the red ocean competition of generalized AI applications, precisely anchoring itself in the "deep waters" of industrial-grade intelligent agents, establishing an absolute monopolistic advantage in this niche segment.
According to Frost & Sullivan data, in 2024, the company held a remarkable 50% market share in China's graph-centric AI intelligent agent market, far exceeding the second place (15%) and third place (10%). This leading position not only helps resist industry competition but also allows it to优先享受 the growth红利 of the赛道. Frost & Sullivan data indicates that the market size for industrial-grade AI solutions in China is expected to grow from approximately RMB 65.4 billion in 2025 to about RMB 286.1 billion in 2029, representing a compound annual growth rate (CAGR) of 44.6%. Behind this astonishing growth rate are three driving forces: accelerated enterprise digital transformation, increasing maturity of AI technology, and rising industry compliance requirements.
More promisingly, the细分赛道 of "graph-centric AI intelligent agents" where Hai Zhi Technology operates is expected to grow even faster than the industry average. Industrial AI intelligent agents integrated with knowledge graphs are projected to surge from RMB 200 million in 2024 to RMB 13.2 billion in 2029, achieving a staggering CAGR of 140.0%.
Looking at the company's performance, its business has entered a period of rapid scaling: total revenue reached RMB 503 million in 2024, with a CAGR of 26.8% from 2022 to 2024. The year-on-year growth rate further accelerated to 38.4% in the first half of 2025, indicating持续强化 growth momentum.
The performance of the core growth engine, the Atlas intelligent agent, is particularly outstanding, becoming a key variable optimizing the revenue structure. This business generated only RMB 9 million in revenue in 2023 but soared to RMB 87 million in 2024. By the first half of 2025, its contribution to total revenue had increased to 28%. This explosive growth not only confirms the commercial potential of graph-model fusion technology but also reflects the market's urgent demand for core business-level AI solutions.
Based on 2024 revenue, Hai Zhi Technology ranks fifth among industrial-grade AI intelligent agent providers in China. However, compared to companies with larger revenue scales, Hai Zhi Technology holds an advantage in the quality of its revenue structure. According to the prospectus, among the two main categories of products and services offered by Hai Zhi Technology, the gross profit margin for Atlas Graph Solutions remained stable at around 35%, while the gross profit margin for Atlas Intelligent Agents increased to approximately 50% in 2024 and 2025.
Judging from the持续合作 of core clients, Hai Zhi Technology's solutions can create tangible benefits for customers, resulting in极强的 customer stickiness and repurchase意愿. This commercialization characteristic of "high contract value + strong payment willingness + pure AI attributes" signifies that Hai Zhi Technology has firmly established itself in the industrial-grade AI market, successfully translating technological value into commercial value and laying a solid foundation for scaled expansion.
Hai Zhi Technology has successfully passed the listing hearing of the Hong Kong Stock Exchange, entering a critical stage in its冲刺 to become the "first AI de-hallucination stock." It is understood that the proceeds from this IPO will be primarily allocated towards strengthening graph-model fusion technology, optimizing the Atlas intelligent agent, expanding into new scenarios and overseas markets, and strategic investments and acquisitions. This will provide additional resources for R&D and market投入, further consolidating its advantage in the细分赛道.
Within the AI ecosystem dominated by giants, Hai Zhi Technology has charted a differentiated path of vertical deep cultivation. As an independent vendor, it does not directly confront giants like Alibaba and Baidu in the field of general AI but instead focuses on the vertical technical赛道 of graph-model fusion, building an irreplaceable core competitiveness.
In summary, Hai Zhi Technology is a rare contender during the challenging phase of AI industrial implementation. Its core investment value concentrates on three points: first, its unique technological barriers; second, the significant赛道红利; and third, its solid commercialization. In the short term, close attention should be paid to post-listing share price stability and the pace of performance improvement. From a medium to long-term perspective, the company has the potential to grow into a leading enterprise in the field of industrial-grade AI de-hallucination, leveraging its technological monopoly and赛道红利 to enjoy the benefits of both valuation and performance appreciation.
Comments